The ClickHouse MCP runs as a **local stdio server** connecting to your ClickHouse instance: - **Database Credentials**: Your connection credentials are stored in environment variables - **Query Execution**: The AI can execute any SQL query your user has permissions for - **Network Access**: Connects to your ClickHouse host (cloud or self-hosted) - **TLS Configuration**: Secure connections are configurable via `CLICKHOUSE_SECURE` and `CLICKHOUSE_VERIFY` **Security best practices**: Use a dedicated read-only user for analytics queries, configure proper TLS settings for production databases, set reasonable query timeouts to prevent runaway queries, and restrict access to sensitive columns using ClickHouse's role-based access control.

To start using the ClickHouse MCP: 1. **Get your connection details**: Host, port, username, and password from your ClickHouse instance 2. **Set environment variables**: ```bash export CLICKHOUSE_HOST="your-host.clickhouse.cloud" export CLICKHOUSE_PORT="8443" export CLICKHOUSE_USER="default" export CLICKHOUSE_PASSWORD="your-password" export CLICKHOUSE_SECURE="true" ``` 3. **Install uv** (if needed): `pip install uv` or `brew install uv` 4. **Add to your AI client**: Use `uvx mcp-clickhouse` 5. **Start analyzing**: Ask your AI questions about your analytics data **Example**: "Show me the top 10 pages by view count from the last 24 hours"